An agent for deception detection in discussion basedenvironments

Amos Azaria, Ariella Richardson, Sarit Kraus

Abstract: Autonomous agents can be of assistance in detecting and reducing deception in computerized forums and chat-rooms. We focus on text-based environments where the deceiver is a member of a group which is holding a discussion. Deception detection methods which currently exist for such environments, heavily rely on either audio or visual information. We have developed DIG, an innovative machine learning-based autonomous agent, which joins a group of players as a regular member and assists them in catching a deceiver. We introduce "the pirate game" as a platform for deploying this agent. Our experimental study shows that although humans display difficulty detecting deception, DIG is not only capable of finding a deceptive player, it also helps increase the entire group's success.